HARNESS: Heterogeneous Cloud Computing

The HARNESS project developed a new generation of cloud computing platform that incorporates non-traditional and heterogeneous computational, networking and storage resources into the data centre stack to provide vastly increased performance at lower cost.

To better service workloads in application domains currently not supported by existing data centres, HARNESS has designed and implemented an enhanced cloud platform stack that fully embraces heterogeneous technologies. A key principle is that both specialised and commodity resources are treated as first-class cloud entities. Thus, the platform supports standard VMs and LXC containers but can also be used to instantiate virtualised FPGA resources and PCIe‐based accelerators (e.g. GPGPUs) as well as managed storage volumes and virtual network links with QoS guarantees.

The HARNESS cloud platform has demonstrated its results via extensions to OpenStack and has been deployed on two testbeds: Grid’5000, a large‐scale and versatile testbed spanning France and consisting of 1000s of nodes along with heterogeneous compute and networking technologies; and a smaller testbed at Imperial College London that incorporates virtualised FPGA‐based dataflow technology from Maxeler Technologies.

EU FP7 (2012-2015)
Alexandros Koliousis (Graphcore, UK)
Paolo Costa (Microsoft Research Cambridge)